AI-Based Smart Dustbin for Automated Waste Segregation Using Deep Learning

AI-Based Smart Dustbin for Automated Waste Segregation Using Deep Learning

Albin P Sabu

Department of Computer Science and Engineering College of Engineering, Kottarakkara Kerala, India

Roshin Shibu Thomas

Department of Computer Science and Engineering College of Engineering, Kottarakkara Kerala, India

Prince Jaison

Department of Computer Science and Engineering College of Engineering, Kottarakkara Kerala, India

Shibin B

Department of Computer Science and Engineering College of Engineering Kottarakkara, Kerala, India

Abstract – Maintaining environmental sustainability and in- creasing recycling rates depend heavily on effective waste segre- gation. Manual labor is a major component of traditional garbage sorting techniques, and it is frequently hazardous, slow, and inconsistent. The design and development of an AI-based smart dustbin that can automatically identify, categorize, and separate waste products is shown in this study. A YOLOv8 deep learning model and a Raspberry Pi 5 are used in the systems construction for real-time image-based garbage classification. An ultrasonic sensor senses the presence of a trash item on the input flap and initiates webcam image capturing. The garbage is classified by the trained model as glass, paper, plastic, metal, or biodegradable. Based on the prediction, a stepper motor rotates the internal bin to the appropriate position, and two SG90 servo motors operate the flap to drop the waste into the correct compartment. Additionally, gas and moisture sensors monitor environmental conditions inside the dustbin to ensure hygiene and safety. The proposed system demonstrates reliable performance and provides a practical solution for smart waste management.

1. INTRODUCTION

   Effective waste management has become a major concern due to rapid urbanization and the continuous increase in solid waste generation. Improper waste disposal and lack of segrega- tion at the source result in environmental pollution, inefficient recycling processes, and health risks for sanitation workers. Traditional waste sorting methods mainly depend on manual labor, which is time-consuming, inconsistent, and unsafe when dealing with hazardous materials. These challenges highlight the need for an automated and intelligent waste segregation system.

   Recent advancements in artificial intelligence and computer vision provide practical solutions for automating waste classifi- cation. Deep learning models are capable of analyzing images and identifying objects with high accuracy. When integrated with compact embedded platforms such as Raspberry Pi,

   Ancy Das Y R

   Assistant Professor

   Department of Computer Science and Engineering College of Engineering Kottarakkara

   Kerala, India

   these technologies enable the development of real-time smart systems suitable for small-scale and urban waste management applications.

   In this project, an AI-based Smart Dustbin is designed and implemented to automatically detect, classify, and segregate waste materials. The system uses an ultrasonic sensor to detect when an item is placed on the flap. Once detected, a webcam captures an image of the waste, and the Raspberry Pi 5 processes the image using a trained YOLOv8 model. The model classifies the waste into one of five categories: paper, plastic, metal, biodegradable, or glass.

   Based on the classification result, a stepper motor rotates the internal bin to the appropriate compartment, and two SG90 servo motors control the flap mechanism to dispose of the waste accurately. Additionally, the system integrates an MQ- 35 gas sensor and a moisture sensor to monitor harmful gases and wet conditions inside the dustbin. This combination of in- telligent classification and environmental monitoring enhances both efficiency and hygiene.

   The primary goal of this project is to develop a compact, cost-effective, and reliable automated waste segregation sys- tem that reduces human involvement while improving opera- tional accuracy. By combining deep learning with mechanical automation, the proposed smart dustbin contributes toward sustainable and intelligent waste management solutions.

2. SYSTEM OVERVIEW

   The proposed AI-Based Smart Dustbin is designed to auto- matically detect, classify, and segregate waste materials using deep learning and embedded hardware. The system integrates sensing modules, image processing, motor-driven actuation, and environmental monitoring to achieve fully automated operation.

   The primary objective of the system is to classify waste into five predefined categories: paper, plastic, metal, biodegradable, and glass. The classification is performed using a trained YOLOv8 object detection model deployed on Raspberry Pi

   5. The system operates in real time and does not require continuous human supervision.

   The overall architecture consists of four major subsystems:

   • Detection Subsystem
   • Classification Subsystem
   • Mechanical Segregation Subsystem
   • Environmental Monitoring Subsystem

   Fig. 1. Block Diagram of the Proposed AI-Based Smart Dustbin System

   1. Detection Subsystem

      The detection process begins when a waste item is placed on the input flap. An ultrasonic sensor continuously measures the distance between the sensor and the flap surface. When an object is detected within a predefined threshold distance, the sensor sends a signal to the Raspberry Pi 5. This signal acts as a trigger to initiate the image capture and classification process.

      This mechanism ensures that the system remains energy efficient by processing images only when waste is present.

   2. Classification Subsystem

      Once the ultrasonic sensor confirms waste placement, a webcam connected to the Raspberry Pi captures an image of the object. The captured image is then processed using the YOLOv8 model for classification.

      The model analyzes visual features such as shape, color, and texture to predict the correct waste category. The predicted class determines the subsequent mechanical action of the system.

   3. Mechanical Segregation Subsystem

      After classification, the stepper motor rotates the internal circular bin structure to align the appropriate compartment

      with the disposal flap. Each compartment corresponds to a specific waste category.

      Two SG90 servo motors control the flap mechanism. The flap opens to allow the waste to drop into the aligned com- partment and closes automatically after disposal. The system then resets and waits for the next waste input.

   4. Environmental Monitoring Subsystem

      The system incorporates an MQ-35 gas sensor to detect harmful gases produced from decomposing waste. A moisture sensor (V2) monitors internal wetness levels.

      These sensors continuously track environmental conditions inside the dustbin, improving hygiene and operational safety.

   5. Overall System Workflow

      The complete working sequence of the smart dustbin is as follows:

      1. Waste is placed on the flap.
      2. Ultrasonic sensor detects the object.
      3. Raspberry Pi triggers image capture.
      4. YOLOv8 predicts the waste category.
      5. Stepper motor rotates to the corresponding bin section.</p
      6. Servo motors open the flap.
      7. Waste is deposited into the correct compartment.
      8. Flap closes and the system returns to standby mode.
3. WORKING MECHANISM

   The AI-Based Smart Dustbin operates through a sequential and automated process that integrates sensing, image process- ing, decision-making, and mechanical actuation. The complete working mechanism is explained below.

   1. Step 1: Waste Placement Detection

      The process begins when a user places a waste item on the input flap of the dustbin. An ultrasonic sensor mounted near the flap continuously measures the distance between the sensor and the surface. When an object is detected within a predefined threshold range, the sensor sends a trigger signal to the Raspberry Pi 5. This ensures that the system activates only when waste is present, thereby reducing unnecessary processing.

   2. Step 2: Image Capture

      After receiving the trigger signal, the Raspberry Pi activates the connected webcam to capture an image of the waste item placed on the flap. The captured image is temporarily stored and prepared for classification. Proper lighting conditions improve image clarity and increase prediction accuracy.

   3. Step 3: Waste Classification Using YOLOv8

      The captured image is passed to the trained YOLOv8 deep learning model deployed on the Raspberry Pi 5. The model analyzes visual features such as color, texture, shape, and edges to identify the waste category. The system classifies the waste into one of the following five classes:

      • Paper
      • Plastic
      • Metal
      • Biodegradable
      • Glass

        The predicted class label is then used to control the me- chanical segregation system.

   4. Step 4: Bin Position Alignment

      Once the waste category is identified, the Raspberry Pi sends control signals to the stepper motor. The stepper motor rotates the internal circular bin structure to align the appropriate com- partment directly below the disposal flap. Each compartment corresponds to a specific waste category.

      The use of a stepper motor ensures precise angular posi- tioning and accurate alignment.

   5. Step 5: Flap Operation and Waste Disposal

      After alignment is confirmed, two SG90 servo motors oper- ate the flap mechanism. The flap opens automatically, allowing the waste item to fall into the correctly aligned compartment. After disposal, the servo motors return the flap to its closed position to prevent odor leakage and maintain hygiene.

   6. Step 6: Environmental Monitoring

      While the segregation process occurs, the environmental monitoring subsystem continuously operates in the back- ground. The MQ-35 gas sensor detects harmful gases that may be produced from decomposing waste. Simultaneously, the moisture sensor measures the internal wetness level of the dustbin.

      If abnormal gas concentration or excessive moisture is detected, the system can be extended to generate alerts or notifications in future upgrades.

   7. Step 7: System Reset and Standby Mode

   After completing the disposal process, the system returns to standby mode. The ultrasonic sensor resumes continuous monitoring for the next waste input. This cyclic process enables real-time and uninterrupted automated operation.

   The integration of intelligent classification, motor-based ac- tuation, and environmental sensing ensures accurate, hygienic, and efficient waste segregation.

4. HARDWARE COMPONENTS

   The proposed AI-Based Smart Dustbin integrates multi- ple hardware components to achieve automated waste detec- tion, classification, segregation, and environmental monitoring. Each component plays a specific role in ensuring accurate and efficient system operation.

   1. Raspberry Pi 5

      Raspberry Pi 5 serves as the central processing unit of the system. It performs image processing, runs the trained YOLOv8 model, and controls motor operations based on classification results. The Raspberry Pi receives input signals

      from sensors, processes the captured image, and generates output control signals for the stepper motor and servo motors. Its compact size and sufficient computational capability make it suitable for edge-based real-time inference.

   2. Ultrasonic Sensor

      The ultrasonic sensor is used to detect the presence of waste on the input flap. It operates by transmitting ultrasonic sound waves and measuring the time taken for the echo to return. When an object is detected within a predefined distance range, the sensor sends a trigger signal to the Raspberry Pi. This ensures that image processing is initiated only when waste is present, improving energy efficiency.

   3. Webcam Module

      A USB webcam is connected to the Raspberry Pi to capture images of the waste item. The clarity and quality of the captured image directly affect the classification accuracy of the YOLOv8 model. The camera captures real-time images when triggered by the ultrasonic sensor and sends them to the Raspberry Pi for processing.

   4. SG90 Servo Motors

      Two SG90 servo motors are used to operate the flap mechanism of the dustbin. Servo motors are chosen because of their precise angular control and fast response. When the appropriate bin compartment is aligned, the servo motors rotate to open the flap, allowing the waste to fall into the correct section. After disposal, the servos return the flap to the closed position.

   5. Stepper Motor

      A stepper motor is used to rotate the internal circular waste bin structure. Each compartment inside the bin corresponds to a specific waste category. The stepper motor provides ac- curate rotational movement, ensuring precise alignment of the selected compartment beneath the disposal flap. This precision is essential for correct segregation.

   6. MQ-35 Gas Sensor

      The MQ-35 gas sensor is integrated to detect harmful gases produced by decomposing waste materials. It continuously monitors gas concentration levels inside the dustbin. Moni- toring gas levels improves safety and hygiene by identifying potential hazardous conditions.

   7. Moisture Sensor (V2)

      The moisture sensor is used to detect the presence of wet waste and measure internal humidity levels. This sensor helps in monitoring damp conditions that may lead to odor formation or bacterial growth. It enhances the environmental monitoring capability of the system.

   8. Power Supply Unit

   A stable power supply unit is required to provide regulated voltage to the Raspberry Pi, sensors, and motor drivers. Proper power management ensures reliable operation and prevents hardware malfunction due to voltage fluctuations.

5. METHODOLOGY

   The methodology of the proposed AI-Based Smart Dustbin involves dataset preparation, model training, system integra- tion, and deployment on embedded hardware. The entire process is designed to achieve accurate real-time waste clas- sification and automated segregation.

   1. Dataset Preparation

      A labeled image dataset containing five waste categories was prepared for training the YOLOv8 model. The categories include:

      • Paper
      • Plastic
      • Metal
      • Biodegradable
      • Glass

        The dataset consits of multiple images collected under dif- ferent lighting conditions and backgrounds to improve model generalization. Images were resized to a uniform resolution suitable for YOLOv8 training. Data augmentation techniques such as rotation, flipping, scaling, and brightness adjustment were applied to increase dataset diversity and reduce overfit- ting.

        Each image was manually annotated using bounding boxes to identify the waste object and assign the correct class label.

   2. Model Selection

      YOLOv8 was selected as the object detection model due to its high detection accuracy and efficient real-time performance. YOLO (You Only Look Once) performs object localization and classification in a single forward pass, making it suitable for embedded systems like Raspberry Pi.

   3. Model Training

      The training process was carried out using the prepared dataset. The dataset was divided into training and validation sets. During training, the model learned to identify distinguish- ing features of each waste category based on shape, texture, and color characteristics.

      Key training parameters included:

      • Multiple training epochs
      • Optimized learning rate
      • Batch size selection based on hardware capability
        • Loss function monitoring for performance improvement

          Model performance was evaluated using validation accuracy and loss metrics. The training process continued until the

          model achieved stable and satisfactory accuracy.

   4. Model Optimization and Export

      After training, the best-performing model weights were selected and exported for deployment. Since Raspberry Pi has limited computational resources compared to high-end GPUs, the model was optimized to ensure smooth real-time inference. The optimized model was converted into a format compat-

      ible with Raspberry Pi for edge deployment.

   5. System Integration

      The deployed YOLOv8 model was integrated with the Raspberry Pi 5. The overall working process was implemented as follows:

      1. Ultrasonic sensor detects waste placement.
      2. Raspberry Pi activates webcam to capture image.
      3. Captured image is processed by YOLOv8.
      4. Predicted class determines motor control action.
      5. Stepper motor aligns correct bin compartment.
      6. Servo motors open and close the flap.

      Python scripts were developed to handle sensor input, image processing, model inference, and motor control operations. GPIO pins were configured to communicate with motors and sensors.

   6. Environmental Monitoring Integration

      The MQ-35 gas sensor and moisture sensor were integrated with the Raspberry Pi using appropriate GPIO connections. Continuous sensor readings were monitored to detect harm- ful gases and internal moisture levels. These values can be extended for alert generation in future enhancements.

   7. Testing and Validation

   The system was tested using various waste items from all five categories. Multiple trials were conducted to verify clas- sification accuracy and mechanical response reliability. The synchronization between detection, classification, and motor control was carefully calibrated to ensure smooth operation.

   The methodology ensures that the system operates in a fully automated and real-time manner with minimal human intervention.

6. IMPLEMENTATION

   The implementation phase focuses on integrating hardware components with the trained deep learning model to achieve real-time automated waste segregation. The system combines embedded programming, sensor interfacing, motor control, and AI-based inference on Raspberry Pi 5.

   1. Hardware Setup

      All hardware components were assembled into a compact dustbin structure. The ultrasonic sensor was mounted near the input flap to detect waste placement. The webcam was positioned above the flap to capture clear images of the waste item.

      The stepper motor was connected to the rotating internal bin structure to ensure accurate alignment of compartments. Two SG90 servo motors were fixed to operate the flap mechanism. The MQ-35 gas sensor and moisture sensor were installed inside the dustbin to continuously monitor environmental conditions.

      Proper wiring and stable power connections were ensured to avoid voltage drops and signal interference.

   2. GPIO Configuration

      The Raspberry Pi 5 GPIO pins were configured to interface with sensors and motors. The ultrasonic sensor used trigger and echo pins for distance measurement. Servo motors were controlled using PWM (Pulse Width Modulation) signals to achieve precise angular movement.

      The stepper motor was controlled through a motor driver module, which allowed controlled rotational steps for accurate bin positioning. Sensor data from the MQ-35 and moisture sensor were read periodically through GPIO inputs.

   3. Circuit Diagram

      The complete hardware circuit diagram of the proposed system is shown in Fig. 2. The diagram illustrates the electrical connections between Raspberry Pi 5, ultrasonic sensor, servo motors, stepper motor driver, gas sensor, and moisture sensor.

      Fig. 2. Circuit Diagram of AI-Based Smart Dustbin

   4. Software Implementation

      Python was used as the primary programming language due to its compatibility with Raspberry Pi and support for deep learning frameworks. The following software compo- nents were integrated:

      • OpenCV for image capture and preprocessing
      • YOLOv8 framework for object detection
      • RPi.GPIO library for hardware interfacing
      • Motor control scripts for servo and stepper operations

        A main control script was developed to coordinate all operations. The program continuously monitors the ultrasonic sensor. When waste is detected, the image capture function is triggered. The captured image is then passed to the YOLOv8 model for classification.

   5. Motor Control Logic

      Based on the predicted waste category, predefined motor positions were assigned to each bin compartment. The stepper motor rotates to the corresponding angular position. After alignment, the servo motors open the flap for a fixed duration to allow waste disposal. The flap then closes automatically.

      This controlled timing ensures accurate and smooth segre- gation without mechanical collision or misalignment.

   6. Environmental Monitoring Implementation

      The MQ-35 gas sensor continuously measures gas concen- tration levels inside the bin. The moisture sensor monitors internal wetness. Sensor readings are processed in the back- ground while the classification system operates. This parallel monitoring ensures improved hygiene and safety.

   7. System Testing

   Multiple real-world tests were conducted using different waste materials. Calibration was performed to adjust motor angles and sensor thresholds. The synchronization between de- tection, classification, and mechanical actuation was optimized to reduce delays.

   The final system achieved stable real-time performance with consistent segregation accuracy.

7. Results and Performance Analysis

   The proposed AI-Based Smart Dustbin was evaluated under real-time operating conditions to assess classification accuracy, mechanical reliability, and overall system responsiveness.Mul- tiple test trials were conducted using different waste items across all five categories.

   1. Classification Performance

      The YOLOv8 model was tested using sample waste items from each category. The classification performance was eval- uated based on prediction accuracy and inference time.

      TABLE I

      Classification Performance of YOLOv8 Model

      Waste Category	Test Samples	Accuracy (%)
      Paper Plastic Metal Biodegradable Glass	20 20 20 20 20	90 92 88 91 89
      Overall Accuracy	100	90

      The model demonstrated stable performance across all categories. Slight variations in accuracy were observed due to lighting conditions and similarities between certain waste materials.

   2. Inference Time Analysis

      The average inference time of the YOLOv8 model on Raspberry Pi 5 was approximately 0.7 to 1.0 seconds per image. This processing speed is suitable for real-time operation in small-scale waste segregation systems.

   3. Mechanical Performance

      The stepper motor accurately rotated the bin to the required compartment in each test trial. The SG90 servo motors re- liably opened and closed the flap mechanism without delay. The synchronization between classification output and motor response was successfully maintained.

   4. Environmental Monitoring Performance

      The MQ-35 gas sensor effectively detected gas presence inside the dustbin. The moisture sensor provided stable read- ings for wet waste conditions. Continuous monitoring ensured improved hygiene and operational safety.

   5. System Reliability

   During repeated testing cycles, the system demonstrated consistent performance without major hardware or software failure. Minor misclassifications occurred primarily under poor lighting or when mixed waste materials were presented.

   Overall, the AI-Based Smart Dustbin achieved reliable real- time waste segregation with satisfactory classification accuracy and mechanical stability.

8. IMPLEMENTATION DETAILS

   This section describes the GPIO pin configuration, hardware interfacing, and execution flow of the AI-Based Smart Dustbin system implemented using Raspberry Pi 5.

   A. GPIO Pin Configuration

   The Raspberry Pi 5 GPIO pins were configured to inter- face with sensors and motor drivers. Table II shows the pin allocation used in the system.

   TABLE II

   GPIO Pin Configuration

   Component	GPIO Pin	Purpose
   Ultrasonic Trigger	GPIO 23	Distance Signal Output
   Ultrasonic Echo	GPIO 24	Distance Signal Input
   Servo Motor 1	GPIO 17	Flap Control
   Servo Motor 2	GPIO 27	Flap Control
   Stepper Motor IN1	GPIO 5	Rotation Control
   Stepper Motor IN2	GPIO 6	Rotation Control
   Stepper Motor IN3	GPIO 13	Rotation Control
   Stepper Motor IN4	GPIO 19	Rotation Control
   MQ-35 Gas Sensor	GPIO 22	Gas Detection Input
   Moisture Sensor	GPIO 18	Moisture Detection Input

   position. Upon receiving the classification result, the Rasp- berry Pi activates the stepper motor to rotate to the required compartment.

   D. Control Flow of the System

   The complete execution flow of the system is described below:

   1. Continuously monitor ultrasonic sensor.
   2. If object detected, capture image using webcam.
   3. Preprocess image and pass it to YOLOv8 model.
   4. Receive predicted waste class.
   5. Map predicted class to predefined bin position.
   6. Rotate stepper motor to align bin compartment.
   7. Activate servo motors to open flap.
   8. Allow waste to fall into compartment.
   9. Close flap after fixed delay.
   10. Continue environmental monitoring using gas and mois- ture sensors.
   11. Return to standby mode.

   E. Software Integration

   All components were integrated using a centralized Python control script. The program follows an event-driven struc- ture, ensuring synchronized communication between detec- tion, classification, and mechanical actuation. Exception han- dling mechanisms were included to prevent system crashes due to unexpected sensor readings.

   This implementation ensures reliable real-time performance and smooth coordination between hardware and software components.

   The GPIO pins were configured using the RPi.GPIO library in Python. Servo motors were controlled using PWM signals to achieve accurate angular movement. The stepper motor was controlled through sequential digital output signals using a motor driver module.

   B. Ultrasonic Sensor Interfacing

   The ultrasonic sensor operates by sending a trigger pulse and measuring the time delay of the echo signal. The distance is calculated using the time-of-flight principle. When the measured distance falls below a predefined threshold, the Raspberry Pi initiates the classification process.

   C. Motor Control Implementation

   The two SG90 servo motors were programmed to rotate between 0° and 90° positions to open and close the flap. A time delay was implemented to ensure complete waste disposal before closing the flap.

   The stepper motor was controlled using a step sequence pattern. Each waste category was assigned a specific angular

9. COST ANALYSIS

   Along with system performance, the overall implementa- tion cost is an important factor when designing practical smart waste management solutions. The proposed AI-Based Smart Dustbin was developed with the goal of keeping the system affordable while still providing reliable automated waste segregation. Since the system is intended for use in environments such as educational institutions, office spaces, and small public areas, the cost of hardware components must remain reasonable to support real-world deployment.

   To achieve this objective, the system makes use of com- monly available embedded hardware components including a Raspberry Pi, sensors, servo motors, a stepper motor, and a webcam module. These components are widely used in embedded and IoT-based applications and are easily available from electronic hardware suppliers. The Raspberry Pi serves as the central processing unit of the system and is responsible for executing the deep learning model, processing captured images, and controlling the motors used for waste segregation. By running the trained YOLOv8 model directly on the Raspberry Pi, the system avoids the need for external cloud- based processing or expensive high- performance computing hardware. This edge-based processing approach helps reduce both installation and operational costs while still aintaining real-time system performance.

   Table III presents the approximate cost of the main hardware components used in the implementation of the proposed smart dustbin system.

   TABLE III

   Estimated Hardware Cost of the Proposed System

   Component	Quantity	Approx. Cost (INR)
   Raspberry Pi 5	1	6500
   USB Webcam	1	800
   Ultrasonic Sensor (HC-SR04)	1	120
   SG90 Servo Motors	2	200
   Stepper Motor with Driver	1	350
   MQ-35 Gas Sensor	1	250
   Moisture Sensor (V2)	1	120
   Power Supply Unit	1	300
   Connecting Wires and Miscella- neous Components		250
   Mechanical Bin Structure	1	400
   Total Estimated Cost		9290

   When compared with industrial waste segregation machines that rely on large conveyor mechanisms and high-performance computing systems, the proposed solution provides a much more affordable alternative. In addition, the use of low-power embedded hardware helps reduce electricity consumption dur- ing operation.

   Another advantage of the proposed design is its modular structure. Individual components can be replaced or upgraded easily without modifying the entire system. This makes the system suitable for deployment in small-scale environments such as schools, universities, offices, and pilot smart city waste management projects.

10. COMPARISON WITH EXISTING METHODS

    To evaluate the effectiveness of the proposed AI-Based Smart Dustbin, a comparative analysis was conducted against commonly used waste classification approaches. The evalua- tion criteria include classification accuracy, real-time capabil- ity, level of automation, and hardware cost requirements.

    TABLE IV

    Comparison with Existing Methods

    Method	Acc (%)	RT	Auto	Cost
    Manual Segregation	70	No	Partial	Low
    Traditional CNN	82	Moderate	Yes	Medium
    MobileNet Model	85	Yes	Yes	Medium
    Proposed Deep Learning System	90	Yes	Yes	Low-Medium

    From Table IV, it can be observed that the proposed deep learning-based system achieves higher classification accuracy compared to traditional CNN-based models and manual seg- regation methods. Unlike manual sorting, the proposed system ensures fully automated operation with minimal human inter- vention.

    Additionally, the deployment on Raspberry Pi 5 enables real- time processing without requiring high-end computational resources. The system maintains a balance between perfor- mance and cost efficiency, making it suitable for practical smart waste management applications.

11. ADVANTAGES OF THE PROPOSED SYSTEM

    The AI-Based Smart Dustbin offers several technical and practical advantages compared to traditional waste segregation methods. The integration of deep learning, embedded systems, and automated mechanical control enhances efficiency, hy- giene, and operational reliability.

    1. Automated Waste Segregation

       The primary advantage of the proposed system is fully automated waste classification and segregation. By utilizing the YOLOv8 deep learning model, the system can accurately identify waste categories without manual intervention. This reduces dependency on human labor and minimizes sorting errors.

    2. Improved Recycling Efficiency

       Accurate segregation at the source significantly improves recycling processes. Proper separation of paper, plastic, metal, biodegradable, and glass waste reduces contamination and in- creases material recovery rates. This contributes to sustainable waste management practices.

    3. Real-Time Operation

       The deployment of the trained model on Raspberry Pi 5 enables real-time image processing and decision-making. The average inference time allows smooth and continuous operation, making the system suitable for practical applications in institutions and small urban setups.

    4. Hygienic and Safe Operation

       The system minimizes direct human contact with waste materials, thereby reducing health risks for sanitation workers. The automatic flap mechanism ensures that waste is disposed of without physical handling.

    5. Environmental Monitoring Capability

       The integration of the MQ-35 gas sensor and moisture sensor enhances safety and hygiene. Continuous monitoring of harmful gases and internal moisture levels helps detect unfavorable conditions inside the dustbin. This feature adds an additional layer of environmental awareness to the system.

    6. Precise Mechanical Control

       The use of a stepper motor ensures accurate alignment of waste compartments, while SG90 servo motors provide precise flap control. This mechanical precision improves segregation reliability and reduces operational errors.

    7. Energy Efficient Edge Processing

       Since the YOLOv8 model runs locally on Raspberry Pi 5, the system does not require continuous cloud connectivity.

       Edge-based processing reduces latency, improves data privacy, and ensures uninterrupted operation even without internet access.

    8. Scalability and Future Expandability

    The modular design of the system allows future upgrades such as IoT-based monitoring dashboards, solar power integra- tion, alert systems, and multi-label waste classification. This scalability makes the system adaptable to larger smart city applications.

    Overall, the proposed AI-Based Smart Dustbin provides an efficient, hygienic, and intelligent solution for automated waste management, contributing toward sustainable and smart environmental practices.

12. Limitations of the Proposed System

    Although the proposed AI-Based Smart Dustbin demon- strates effective automated waste segregation, certain limita- tions were observed during development and testing.

    1. Lighting Sensitivity

       The performance of the YOLOv8 classification model de- pends on image quality. Variations in lighting conditions, shadows, or low illumination may reduce classification accu- racy. Controlled lighting improves performance, but real-world environments may introduce variability.

    2. Limited Dataset Size

       The model was trained on a limited dataset containing five waste categories. If the system encounters unfamiliar waste types or mixed materials, misclassification may occur. Increas- ing dataset size and diversity can improve generalization.

    3. Mixed Waste Classification Challenges

       The current system is designed for single-object classifica- tion. When multiple waste items or mixed materials are placed simultaneously, the model may face difficulty in accurately predicting categories.

    4. Hardware Constraints

       Although Raspberry Pi 5 provides sufficient processing capability, it is still less powerful than high-end GPUs. This may slightly limit inference speed and scalability for more complex models.

    5. Mechanical Wear and Maintenance

       Continuous operation of servo motors and the stepper motor may lead to mechanical wear over time. Regular maintenance is required to ensure smooth and precise functioning.

        

    6. Power Dependency

       The system relies on a stable power supply. Voltage fluctu- ations or power interruptions may affect system reliability and performance.

    7. Sensor Accuracy Limitations

    Gas and moisture sensors provide approximate environmen- tal readings. Extreme environmental conditions may influence sensor accuracy.

    Despite these limitations, the system performs reliably under normal operating conditions and serves as a practical prototype for intelligent waste segregation.

13. FUTURE SCOPE

    The proposed AI-Based Smart Dustbin provides a functional prototype for intelligent waste segregation. However, several improvements and enhancements can be implemented in the future to increase efficiency, scalability, and real-world appli- cability.

    1. Multi-Object and Multi-Label Classification

       The current system classifies a single waste item at a time. Future enhancements can focus on multi-object detection and multi-label classification to handle mixed waste materials more effectively.

    2. Larger and Diverse Dataset

       Expanding the training dataset with more images collected under different environmental conditions can improve model accuracy and robustness. Including additional waste categories will make the system more practical for real-world applica- tions.

    3. Cloud and IoT Integration

       The system can be extended by integrating IoT-based moni- toring dashboards. Real-time data such as waste type statistics, gas levels, and bin fill levels can be uploaded to a cloud platform for remote monitoring and analysis.

    4. Solar Power Integration

       To improve sustainability, the system can be powered using solar panels. This will allow deployment in outdoor and remote locations with reduced dependency on grid electricity.

    5. Alert and Notification System

       Future upgrades may include automated alert systems that notify maintenance personnel when bins are full or when harmful gas levels exceed safe limits.

    6. Improved Mechanical Design

       Enhancing the structural design of the rotating bin and flap mechanism can improve durability and reduce long-term mechanical wear.

    7. Industrial-Level Deployment

    With further optimization, the system can be scaled for use in schools, offices, public areas, and smart city

    waste management systems.

    These future improvements can transform the current pro- totype into a fully scalable and intelligent waste management solution.

14. RESULTS AND DISCUSSION

    The proposed deep learning-based smart waste segregation system was evaluated under real-time operating conditions to analyze classification accuracy, inference speed, mechanical response, and overall system reliability.

    1. Classification Performance

       The trained YOLOv8 model was tested using multiple samples from five waste categories: paper, plastic, metal, biodegradable, and glass. Each category was tested with 20 different samples collected under varying lighting and back- ground conditions.

       TABLE V

       Classification Performance of the Proposed System

       Waste Category	Test Samples	Correct Predictions	Accuracy (%)
       Paper	20	18	90
       Plastic	20	19	95
       Metal	20	17	85
       Biodegradable	20	18	90
       Glass	20	18	90
       Overall	100	90	90

       As shown in Table V, the system achieved an overall classi- fication accuracy of approximately 90%. The highest accuracy was observed in plastic waste detection, while slightly lower performance was recorded for metal waste due to reflective surface variations affecting image features.

    2. Inference Time Analysis

       The average inference time of the YOLOv8 model on Raspberry Pi 5 was observed to be between 0.7 and 1.0 seconds per image. This processing time is suitable for real- time waste segregation applications in controlled environments such as educational institutions and offices.

    3. Mechanical Performance Evaluation

       The stepper motor demonstrated accurate rotational align- ment for bin positioning. Across repeated trials, the alignment error was negligible. The SG90 servo motors successfully opened and closed the flap mechanism without mechanical delay or collision.

       The average total segregation cycle time, including detec- tion, classification, rotation, and disposal, was approximately 3 4 seconds per waste item.

    4. Environmental Monitoring Analysis

       The MQ-35 gas sensor successfully detected variations in gas concentration when biodegradable waste was introduced. The moisture sensor effectively identified wet waste condi- tions.

       These additional monitoring features enhance hygiene and safety within the system.

    5. Discussion

    The experimental results indicate that the proposed system provides reliable and efficient waste segregation with minimal human intervention. Compared to manual sorting, the auto- mated system significantly reduces health risks and improves segregation consistency.

    Although the classification accuracy is high, certain lim- itations were observed under poor lighting conditions and when mixed waste materials were presented simultaneously. Increasing dataset diversity and implementing multi-object detection can further enhance robustness.

    Overall, the integration of deep learning with embedded hardware demonstrates a practical and scalable solution for intelligent waste management systems.

15. CONCLUSION

This paper presented the design and implementation of an AI-Based Smart Dustbin for automated waste segregation using the YOLOv8 deep learning model deployed on Rasp- berry Pi 5. The system integrates ultrasonic sensing, real-time image classification, motor- driven mechanical segregation, and environmental monitoring to achieve intelligent waste manage- ment.

The proposed solution successfully classifies waste into five categories: paper, plastic, metal, biodegradable, and glass. Based on the predicted class, a stepper motor aligns the appropriate bin compartment while servo motors control the disposal flap. Additionally, the integration of gas and moisture sensors enhances safety and hygiene by continuously monitor- ing internal environmental conditions.

Experimental evaluation demonstrated reliable classification accuracy and stable mechanical performance under normal operating conditions. The system reduces manual intervention, improves segregation efficiency, and promotes hygienic waste handling practices.

Although certain limitations such as lighting sensitivity and dataset constraints were observed, the overall system demonstrates strong potential for small-scale smart waste management applications. With further improvements and scalability enhancements, the proposed system can contribute toward sustainable environmental practices and smart city development.

In conclusion, the AI-Based Smart Dustbin represents a practical and intelligent approach to automated waste segre- gation by combining deep learning, embedded systems, and mechanical automation into a compact and efficient solution.

REFERENCES

1. J. Redmon, S Divvala, R. Girshick, and A. Farhadi, You Only Look Once: Unified, Real-Time Object Detection, in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779788.
2. G. Jocher, A. Chaurasia, and J. Qiu, YOLOv8 by Ultralytics, Ultralyt- ics, 2023. [Online]. Available: https://github.com/ultralytics/ultralytics
3. I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA, USA: MIT Press, 2016.
4. A. Krizhevsky, I. Sutskever, and G. E. Hinton, ImageNet Classification with Deep Convolutional Neural Networks, in Advances in Neural Information Processing Systems, 2012, pp. 10971105.
5. H. Zhang, L. Wang, and Z. Liu, Garbage Classification Using Deep Learning: A Survey, IEEE Access, vol. 8, pp. 145160, 2020.
6. M. Nahiduzzaman et al., Automated Waste Classification Using Deep Learning Toward Efficient Recycling, Knowledge-Based Systems, vol. 302, 2025.
7. Raspberry Pi Foundation, Raspberry Pi 5 Documentation, 2023. [On- line]. Available: https://www.raspberrypi.org
8. Hanwei Electronics, MQ-35 Gas Sensor Technical Datasheet, 2022.
9. S. S. S. Kumar and R. Kumar, IoT-Based Smart Waste Management System, in Proc. IEEE Int. Conf. Smart Systems and Inventive Tech- nology, 2019.
10. W. Mao, J. Li, and X. Chen, Deep Learning-Based Smart Waste Segregation System for Sustainable Cities, Sustainable Computing: Informatics and Systems, vol. 34, 2022.